DIABETES DIAGNOSIS ANALYSIS¶

19TH JUNE,2023

1 Problem Statement : Diabetes Diagnosis Analysis¶

1.1 Description:

  • Diabetes Mellitus is one of the currently deadliest diseases as it set the foundation for several conditions to develop in the body. Diabetes paves way for conditions such as high blood pressure, hyperlipidaemia, neuropathy, eye problems, kidney failure, among others. All these conditions also lead to other conditions that are also very dangerous to the body, making diabetes worth studied into more details. Between 2000 and 2019, there was a 3% increase in diabetes mortality rates by age. In 2019, diabetes and kidney disease due to diabetes caused an estimated 2 million deaths. In the diagnosis of diabetes, an important biomarker is blood glucose level, other biomarkers and physical characteristics play important role in the diagnosis as well. This article seeks the investigate the various factors involved in diabetes diagnosis to reveal new other more important features that can be used to detect the presence of diabetes and also to develop a Machine Learning Model to determine the diabetic state of a patient when their information is inputed.

2 1. Importing Libraries¶

In [1]:
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import mean_absolute_error
import warnings
warnings.filterwarnings('ignore')

2 The DataSets¶

2.1 DataSets Information¶

Column attribute Description Patient number Identifies patients by number Cholesterol Total cholesterol Glucose Fasting blood sugar HDL HDL or good cholesterol Chol/HDL Ratio of total cholesterol to good cholesterol. Desirable result is < 5 Age All adult African Americans Gender 162 males, 228 females Height In inches Weight In pounds (lbs) BMI 703 x weight (lbs)/ [height(inches]2 Systolic BP The upper number of blood pressure Diastolic BP The lower number of blood pressure Waist Measured in inches Hip Measured in inches Waist/hip Ratio is possibly a stronger risk factor for heart disease than BMI Diabetes Yes (60), No (330)

This dataset is a modified data from VanderBilt but the original dataset is from a study of rural African American in Virginia. The dataset already has 13 patients dropped due to missing data. It also has two features added, being Chol/HDL and Waist/hip.

2.2 Reading Datasets¶

In [2]:
df = pd.read_csv('\\Users\\Daniel Appiah\\Desktop\\Diabetes_Classification.csv', index_col='Patient number')
df.head()
Out[2]:
Cholesterol Glucose HDL Chol Chol/HDL ratio Age Gender Height Weight BMI Systolic BP Diastolic BP waist hip Waist/hip ratio Diabetes Unnamed: 16 Unnamed: 17
Patient number
1 193 77 49 3.9 19 female 61 119 22.5 118 70 32 38 0.84 No diabetes 6.0 6.0
2 146 79 41 3.6 19 female 60 135 26.4 108 58 33 40 0.83 No diabetes NaN NaN
3 217 75 54 4.0 20 female 67 187 29.3 110 72 40 45 0.89 No diabetes NaN NaN
4 226 97 70 3.2 20 female 64 114 19.6 122 64 31 39 0.79 No diabetes NaN NaN
5 164 91 67 2.4 20 female 70 141 20.2 122 86 32 39 0.82 No diabetes NaN NaN

2.3 Data Exploration¶

In [3]:
df.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 390 entries, 1 to 390
Data columns (total 17 columns):
 #   Column           Non-Null Count  Dtype  
---  ------           --------------  -----  
 0   Cholesterol      390 non-null    int64  
 1   Glucose          390 non-null    int64  
 2   HDL Chol         390 non-null    int64  
 3   Chol/HDL ratio   390 non-null    float64
 4   Age              390 non-null    int64  
 5   Gender           390 non-null    object 
 6   Height           390 non-null    int64  
 7   Weight           390 non-null    int64  
 8   BMI              390 non-null    float64
 9   Systolic BP      390 non-null    int64  
 10  Diastolic BP     390 non-null    int64  
 11  waist            390 non-null    int64  
 12  hip              390 non-null    int64  
 13  Waist/hip ratio  390 non-null    float64
 14  Diabetes         390 non-null    object 
 15  Unnamed: 16      1 non-null      float64
 16  Unnamed: 17      1 non-null      float64
dtypes: float64(5), int64(10), object(2)
memory usage: 54.8+ KB

All features here are numeric except the column named Diabetes.

In [4]:
df.describe()
Out[4]:
Cholesterol Glucose HDL Chol Chol/HDL ratio Age Height Weight BMI Systolic BP Diastolic BP waist hip Waist/hip ratio Unnamed: 16 Unnamed: 17
count 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 390.000000 1.0 1.0
mean 207.230769 107.338462 50.266667 4.524615 46.774359 65.951282 177.407692 28.775641 137.133333 83.289744 37.869231 42.992308 0.881385 6.0 6.0
std 44.666005 53.798188 17.279069 1.736634 16.435911 3.918867 40.407824 6.600915 22.859528 13.498192 5.760947 5.664342 0.073212 NaN NaN
min 78.000000 48.000000 12.000000 1.500000 19.000000 52.000000 99.000000 15.200000 90.000000 48.000000 26.000000 30.000000 0.680000 6.0 6.0
25% 179.000000 81.000000 38.000000 3.200000 34.000000 63.000000 150.250000 24.100000 122.000000 75.000000 33.000000 39.000000 0.830000 6.0 6.0
50% 203.000000 90.000000 46.000000 4.200000 44.500000 66.000000 173.000000 27.800000 136.000000 82.000000 37.000000 42.000000 0.880000 6.0 6.0
75% 229.000000 107.750000 59.000000 5.400000 60.000000 69.000000 200.000000 32.275000 148.000000 90.000000 41.000000 46.000000 0.930000 6.0 6.0
max 443.000000 385.000000 120.000000 19.300000 92.000000 76.000000 325.000000 55.800000 250.000000 124.000000 56.000000 64.000000 1.140000 6.0 6.0
In [5]:
df.corr()
Out[5]:
Cholesterol Glucose HDL Chol Chol/HDL ratio Age Height Weight BMI Systolic BP Diastolic BP waist hip Waist/hip ratio Unnamed: 16 Unnamed: 17
Cholesterol 1.000000 0.158102 0.193162 0.475927 0.247333 -0.063601 0.062359 0.091695 0.207741 0.166241 0.134038 0.093364 0.091847 NaN NaN
Glucose 0.158102 1.000000 -0.158302 0.282210 0.294392 0.098052 0.190358 0.129286 0.162777 0.020262 0.222336 0.138223 0.185117 NaN NaN
HDL Chol 0.193162 -0.158302 1.000000 -0.681867 0.028210 -0.087238 -0.291883 -0.241860 0.031807 0.078342 -0.276697 -0.223837 -0.158777 NaN NaN
Chol/HDL ratio 0.475927 0.282210 -0.681867 1.000000 0.163201 0.081162 0.278812 0.228407 0.115505 0.038242 0.313262 0.208902 0.243329 NaN NaN
Age 0.247333 0.294392 0.028210 0.163201 1.000000 -0.082229 -0.056784 -0.009164 0.453417 0.068649 0.150585 0.004675 0.275188 NaN NaN
Height -0.063601 0.098052 -0.087238 0.081162 -0.082229 1.000000 0.255389 -0.259589 -0.040704 0.043617 0.057447 -0.095906 0.252548 NaN NaN
Weight 0.062359 0.190358 -0.291883 0.278812 -0.056784 0.255389 1.000000 0.860147 0.097497 0.166477 0.847766 0.826985 0.250461 NaN NaN
BMI 0.091695 0.129286 -0.241860 0.228407 -0.009164 -0.259589 0.860147 1.000000 0.121408 0.145304 0.810701 0.881728 0.100873 NaN NaN
Systolic BP 0.207741 0.162777 0.031807 0.115505 0.453417 -0.040704 0.097497 0.121408 1.000000 0.603662 0.210934 0.155321 0.137871 NaN NaN
Diastolic BP 0.166241 0.020262 0.078342 0.038242 0.068649 0.043617 0.166477 0.145304 0.603662 1.000000 0.165846 0.143898 0.077918 NaN NaN
waist 0.134038 0.222336 -0.276697 0.313262 0.150585 0.057447 0.847766 0.810701 0.210934 0.165846 1.000000 0.835177 0.514177 NaN NaN
hip 0.093364 0.138223 -0.223837 0.208902 0.004675 -0.095906 0.826985 0.881728 0.155321 0.143898 0.835177 1.000000 -0.037664 NaN NaN
Waist/hip ratio 0.091847 0.185117 -0.158777 0.243329 0.275188 0.252548 0.250461 0.100873 0.137871 0.077918 0.514177 -0.037664 1.000000 NaN NaN
Unnamed: 16 NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
Unnamed: 17 NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN

2.4 Handling Missing Data¶

In [6]:
df.isnull().sum()
Out[6]:
Cholesterol          0
Glucose              0
HDL Chol             0
Chol/HDL ratio       0
Age                  0
Gender               0
Height               0
Weight               0
BMI                  0
Systolic BP          0
Diastolic BP         0
waist                0
hip                  0
Waist/hip ratio      0
Diabetes             0
Unnamed: 16        389
Unnamed: 17        389
dtype: int64

The data has no missing values except for the last 2 unnamed features, which will be dropped due to the large number of null values.

In [7]:
df= df.drop(['Unnamed: 16','Unnamed: 17'], axis=1)
df.columns
Out[7]:
Index(['Cholesterol', 'Glucose', 'HDL Chol', 'Chol/HDL ratio', 'Age', 'Gender',
       'Height', 'Weight', 'BMI', 'Systolic BP', 'Diastolic BP', 'waist',
       'hip', 'Waist/hip ratio', 'Diabetes'],
      dtype='object')

2.5 Categorical Features¶

In [8]:
unique_values= df.apply(lambda x:x.nunique())
unique_values
Out[8]:
Cholesterol        153
Glucose            116
HDL Chol            75
Chol/HDL ratio      69
Age                 68
Gender               2
Height              22
Weight             139
BMI                193
Systolic BP         71
Diastolic BP        56
waist               30
hip                 32
Waist/hip ratio     39
Diabetes             2
dtype: int64

From this and a critical look at the data, the 'Gender' and 'Diabetes' columns are nominal features and must be changed to categorical data type.

In [9]:
df['Gender'] =df['Gender'].astype('category')
df['Diabetes'] =df['Diabetes'].astype('category')
df['Diabetes'].dtypes
Out[9]:
CategoricalDtype(categories=['Diabetes', 'No diabetes'], ordered=False)

3. Data Visualization¶

3.1 Visualising the distribution of the data¶

Cholesterol level¶

In [10]:
plt.figure(figsize=(5,4))
sns.distplot(df.Cholesterol)
plt.title('Cholesterol Level Distribution(mg/dL)', fontdict={'fontweight': 'bold'})
#Classifying the cholesterol level.
desirable = df[df.Cholesterol < 200].count()[0]
borderline = df[(df.Cholesterol >= 200) & (df.Cholesterol < 240)].count()[0]
high = df[df.Cholesterol >= 240].count()[0]

print(f'Desirable Range (below 200 mg/dL): {desirable} patients.')
print(f'Borderline High Range (200-239 mg/dL): {borderline} patients.')
print(f'High Range (240 mg/dL and above): {high} patients.')

Desirable Range (below 200 mg/dL): 184 patients.
Borderline High Range (200-239 mg/dL): 130 patients.
High Range (240 mg/dL and above): 76 patients.

Glucose level¶

In [11]:
plt.figure(figsize=(5,4))
sns.distplot(df.Glucose)
plt.title('Glucose Level Distribution(mg/dL) ',fontdict={'fontweight': 'bold'})

desirable = df[df.Glucose < 100].count()[0]
borderline = df[(df.Glucose >= 100) & (df.Glucose < 125)].count()[0]
diabetes = df[df.Glucose >= 126].count()[0]

print(f'Normal Range (below 100 mg/dL): {desirable} patients.')
print(f'Prediabetes Range (100-125 mg/dL): {borderline} patients.')
print(f'Diabetes Range (126 mg/dL and above): {diabetes} patients.')

Normal Range (below 100 mg/dL): 260 patients.
Prediabetes Range (100-125 mg/dL): 69 patients.
Diabetes Range (126 mg/dL and above): 60 patients.

High density Lipoprotein¶

In [12]:
plt.figure(figsize=(5,4))
sns.distplot(df['HDL Chol'])
plt.title('High Density Lipoprotein Cholesterol Level Distribution(mg/dL)',fontdict={'fontweight': 'bold'})

desirable_men = df[(df['HDL Chol'] >= 40) & (df['HDL Chol'] < 60) ].count()[0]
desirable_women = df[(df['HDL Chol'] >= 50) & (df['HDL Chol'] < 60)].count()[0]
low_men = df[df['HDL Chol'] <40 ].count()[0]
low_women= df[df['HDL Chol'] <50 ].count()[0]
high = df[df['HDL Chol'] >= 60].count()[0]

print(f'Low Range for men (below 40 mg/dL): {low_men} patients.')
print(f'Low Range for women (below 50 mg/dL): {low_women} patients.')
print(f'Normal HDL Range for men (40-59 mg/dL): {desirable_men} patients.')
print(f'Normal HDL Range for women(50-59 mg/dL): {desirable_women} patients.')
print(f'High HDL Range(60 mg/dL and above): {high} patients.')

Low Range for men (below 40 mg/dL): 108 patients.
Low Range for women (below 50 mg/dL): 226 patients.
Normal HDL Range for men (40-59 mg/dL): 189 patients.
Normal HDL Range for women(50-59 mg/dL): 71 patients.
High HDL Range(60 mg/dL and above): 93 patients.

Age¶

In [13]:
plt.figure(figsize=(5,4))
sns.distplot(df.Age)
plt.title('Age Distribution(yrs)',fontdict={'fontweight': 'bold'})
Out[13]:
Text(0.5, 1.0, 'Age Distribution(yrs)')

Gender¶

In [14]:
plt.figure(figsize=(3,7))
sns.histplot(df.Gender)
plt.title('Gender Distribution',fontdict={'fontweight': 'bold'})
Out[14]:
Text(0.5, 1.0, 'Gender Distribution')

Height¶

In [15]:
plt.figure(figsize=(5,4))
sns.distplot(df.Height)
plt.title('Height Distribution(inches)',fontdict={'fontweight': 'bold'})
Out[15]:
Text(0.5, 1.0, 'Height Distribution(inches)')

Weight¶

In [16]:
plt.figure(figsize=(5,4))
sns.distplot(df.Weight)
plt.title('Weight Distribution(pounds(lbs))',fontdict={'fontweight': 'bold'})
Out[16]:
Text(0.5, 1.0, 'Weight Distribution(pounds(lbs))')

Body Mass Index¶

In [17]:
plt.figure(figsize=(5,4))
sns.distplot(df.BMI)
plt.title('Body Mass Index Distribution(lbs/(in)^2 )',fontdict={'fontweight': 'bold'})

underweight = df[df['BMI'] <18.5].count()[0]
normal_weight = df[(df['BMI'] >= 18.5) & (df['BMI'] < 24.9)].count()[0]
overweight = df[(df['BMI'] >=25.5)  & (df['BMI'] <30 )].count()[0]
obese= df[df['BMI'] >=30 ].count()[0]

print(f'Underweight (below 18.5 lbs/in^2): {underweight} patients.')
print(f'Normal weight (18.5-24.9 lbs/in^2): {normal_weight} patients.')
print(f'Overweight (25-29.9 lbs/in^2): {overweight} patients.')
print(f'Obese (30 lbs/in^2 and above): {obese} patients.')

Underweight (below 18.5 lbs/in^2): 9 patients.
Normal weight (18.5-24.9 lbs/in^2): 106 patients.
Overweight (25-29.9 lbs/in^2): 107 patients.
Obese (30 lbs/in^2 and above): 150 patients.

The BMI classes are as follows: underweight (BMI less than 18.5), normal weight (BMI between 18.5 and 24.9), overweight (BMI between 25 and 29.9), obesity (Class I, BMI between 30 and 34.9), obesity (Class II, BMI between 35 and 39.9), and obesity (Class III, BMI 40 or higher)

Systolic Blood Pressure¶

In [18]:
plt.figure(figsize=(5,4))
sns.distplot(df['Systolic BP'])
plt.title('Systolic Blood Pressure Distribution(mmHg)',fontdict={'fontweight': 'bold'})
Out[18]:
Text(0.5, 1.0, 'Systolic Blood Pressure Distribution(mmHg)')

The systolic blood pressure (SBP) categories are as follows: normal (SBP below 120 mmHg but not up to 90mmHg), elevated (SBP between 120 and 129 mmHg), hypertension stage 1 (SBP between 130 and 139 mmHg), hypertension stage 2 (SBP 140 mmHg or higher), and hypertensive crisis (SBP higher than 180 mmHg and/or diastolic blood pressure (DBP) higher than 120 mmHg). This data has few portion below 120mmHg and the rest above.

Diastolic Blood Pressure¶

In [19]:
plt.figure(figsize=(5,4))
sns.distplot(df['Diastolic BP'])
plt.title('Diastolic Blood Pressure Distribution(mm/Hg)',fontdict={'fontweight': 'bold'})
Out[19]:
Text(0.5, 1.0, 'Diastolic Blood Pressure Distribution(mm/Hg)')

The diastolic blood pressure (DBP) categories are as follows: normal (DBP below 80 mmHg b ut not up to 60mmHg), elevated (DBP between 80 and 89 mmHg), hypertension stage 1 (DBP between 90 and 99 mmHg), hypertension stage 2 (DBP 100 mmHg or higher), and hypertensive crisis (DBP higher than 120 mmHg and/or systolic blood pressure (SBP) higher than 180 mmHg).

Waist size¶

In [20]:
plt.figure(figsize=(5,4))
sns.distplot(df['waist'])
plt.title('Waist Size Distribution(inches)',fontdict={'fontweight': 'bold'})
Out[20]:
Text(0.5, 1.0, 'Waist Size Distribution(inches)')

Hip Size¶

In [21]:
plt.figure(figsize=(5,4))
sns.distplot(df['hip'])
plt.title('Hip Size Distribution(inches)',fontdict={'fontweight': 'bold'})
Out[21]:
Text(0.5, 1.0, 'Hip Size Distribution(inches)')

4 Correlation Analysis¶

4.1 Categorical Feature Analysis¶

In [22]:
from sklearn.preprocessing import LabelEncoder


label_encoder = LabelEncoder()
df['Diabetes'] = label_encoder.fit_transform(df['Diabetes'])
df['Gender'] = label_encoder.fit_transform(df['Gender'])
label_mapping = {0: 1, 1: 0}
df['Diabetes']=df['Diabetes'].map(label_mapping)

df.head()
Out[22]:
Cholesterol Glucose HDL Chol Chol/HDL ratio Age Gender Height Weight BMI Systolic BP Diastolic BP waist hip Waist/hip ratio Diabetes
Patient number
1 193 77 49 3.9 19 0 61 119 22.5 118 70 32 38 0.84 0
2 146 79 41 3.6 19 0 60 135 26.4 108 58 33 40 0.83 0
3 217 75 54 4.0 20 0 67 187 29.3 110 72 40 45 0.89 0
4 226 97 70 3.2 20 0 64 114 19.6 122 64 31 39 0.79 0
5 164 91 67 2.4 20 0 70 141 20.2 122 86 32 39 0.82 0

4.2 Visualizing the correlation among the various features of the data¶

In [23]:
data_correlation= df.corr()
plt.figure(figsize= (7,5))
plt.title('Correlation Heatmap', fontdict= {'fontweight':'bold'})
sns.heatmap(data_correlation,annot=True,fmt='.0%' , annot_kws= {'fontsize': 7})
Out[23]:
<AxesSubplot: title={'center': 'Correlation Heatmap'}>
In [24]:
#cat= df['Diabetes_encoded','Gender_encoded']
#data= df.drop(cat)
plt.figure(figsize= (15,15))
sns.pairplot(df)
Out[24]:
<seaborn.axisgrid.PairGrid at 0x282415ddd0>
<Figure size 1500x1500 with 0 Axes>

Let us take a look at features with very good correlation.

In [25]:
correlation_matrix = df.corr()

high_correlation_pairs = []
for i in range(len(correlation_matrix.columns)):
    for j in range(i + 1, len(correlation_matrix.columns)):
        if abs(correlation_matrix.iloc[i, j]) > 0.30:
            pair = (correlation_matrix.columns[i], correlation_matrix.columns[j])
            high_correlation_pairs.append(pair)

print(high_correlation_pairs)

[('Cholesterol', 'Chol/HDL ratio'), ('Glucose', 'Diabetes'), ('HDL Chol', 'Chol/HDL ratio'), ('Chol/HDL ratio', 'waist'), ('Age', 'Systolic BP'), ('Age', 'Diabetes'), ('Gender', 'Height'), ('Gender', 'Waist/hip ratio'), ('Weight', 'BMI'), ('Weight', 'waist'), ('Weight', 'hip'), ('BMI', 'waist'), ('BMI', 'hip'), ('Systolic BP', 'Diastolic BP'), ('waist', 'hip'), ('waist', 'Waist/hip ratio')]

The pairs above show features that have correlation coeffient above plus or minus 0.3 .

4.3 Visualizing correlation between features and target(Diabetes column)¶

In [26]:
correlation_matrix = df.corr()
target_correlation = correlation_matrix['Diabetes']  
target_correlation = target_correlation.abs().sort_values(ascending=True)
target_correlation = target_correlation.drop('Diabetes')
plt.figure(figsize=(5, 4))
plt.barh(target_correlation.index, target_correlation.values)
plt.xlabel('Correlation')
plt.ylabel('Features')
plt.title('Correlation between Features and Diabetes')
plt.show()

From the graph we can see that glucose level has the highest correlation coeffeicient with Diabetes, then surprisingly, age through to height having the lowest correlation coefficient with Diabetes.

4. 4 Exploring new features¶

Since the age seems to have a good correlation, let us utilise it in exploring new features with higher correlation.

In [27]:
#creating square of the ages feature
age_square= (df.Age)**2
New_features= pd.DataFrame()
New_features['Age_square']= age_square

Waist seems to have a good correlation with Diabetes, I am creating a new feature ((waist)^2/Age) to see how it will correlate with diabetes since a relationship between a person's waist and age can reveals obesity.

In [28]:
Square_Waist_per_Age= df['waist']**2/df['Age']
New_features['Square_Waist/Age']= Square_Waist_per_Age

Creating another feature which is square root of (Glucose level * Weight * (Chol/HDL))

In [29]:
GWC= np.sqrt(df.Glucose * df.Weight * df['Chol/HDL ratio'])
New_features['GWC']= GWC
In [30]:
New_features['Diabetes']= df.Diabetes
New_features['Glucose']=df.Glucose
New_features['BMI']=df.BMI
New_features['Chol/HDL ratio'] =df['Chol/HDL ratio']
New_features['Age']= df.Age

corr_matrix = New_features.corr()
target_correlation = corr_matrix['Diabetes']
target_correlation = target_correlation.abs().sort_values(ascending=True)
target_correlation = target_correlation.drop('Diabetes')
plt.figure(figsize=(5, 4))
color=['orange','grey','grey','orange','grey','orange','grey']
plt.barh(target_correlation.index, target_correlation.values, color=color)
plt.xlabel('Correlation')
plt.ylabel('Features')
plt.title('Correlation between New Features and Diabetes')
plt.show()

print(f'Correlation Coefficient of GWC is  {target_correlation.GWC} whilst that of glucose is {target_correlation.Glucose}')

Correlation Coefficient of GWC is  0.5975911038445685 whilst that of glucose is 0.6890795038664445

From the graph, it could be revealed that the new feature GWC which is the square root of (Weight * Glucose Level * (cholesterol level/HDL level) tends to be of a very high correlation coefficient with Diabetes and can be a very useful factor in diagnosing diabetes even over most of the the traditional factors that have always been used. The feature GWC makes use of glucose level, cholesterol level, weight and HDL level to give a more accurate foresight concerning the likelihood of the patient being diabetic or not. From analysis, GWC above 200 calls for attention since it is more likely patient is diabetic.

5 Prediction Model¶

In [31]:
data= New_features.merge(df)
X = data.drop(columns ='Diabetes')
y = data.Diabetes

X_train,X_test,y_train,y_valid= train_test_split(X,y,test_size=0.2,random_state=0)

my_model= RandomForestClassifier(n_estimators=500 , max_depth= 100, random_state=0)
my_model.fit(X_train,y_train)
predictions = my_model.predict(X_test)

efficiency = my_model.score(X_test, y_valid)
mae= mean_absolute_error(y_valid,predictions)
print('model accuracy:', efficiency)
print('mean absolute error:', mae)

model accuracy: 0.9230769230769231
mean absolute error: 0.07692307692307693

5.2Testing the model¶

In [32]:
feature= np.array([[2400,29,350,83,24,(200/39),45,200,39,0,52,170,122,75,33,42,(33/42)]])
my_model.predict(feature)
Out[32]:
array([0], dtype=int64)

5. 3 Creating an automated diagnosing predictor¶

In [36]:
import pandas as pd
import numpy as np

def get_numeric_input(prompt):
    while True:
        try:
            value = int(input(prompt))
            return value
        except ValueError:
            print("Invalid input. Please enter a numeric value.")

def predict_diabetes():
    age = get_numeric_input('Age of patient: ')
    age_sqr = age ** 2
    waist = get_numeric_input('Waist size in Inches: ')
    weight = get_numeric_input('Weight of patient in lbs: ')
    sqr_waist_per_age = (waist ** 2) / age
    glucose = get_numeric_input('Fasting glucose level in mg/dL: ')
    cholesterol = get_numeric_input('Cholesterol level in mg/dL: ')
    HDL = get_numeric_input('HDL Cholesterol level in mg/dL: ')
    GWC = np.sqrt(glucose * weight * (cholesterol / HDL))
    while True:
        try:
            gender = int(input("Gender (Input '1' for male or '0' for female): "))
            gender = str(gender)
            if gender not in ('0', '1'):
                raise ValueError
            break
        except ValueError:
            print("Invalid input. Please enter either '0' or '1'.")
    height = get_numeric_input('Height of patient in inches: ')
    BMI = 703 * (weight / (height ** 2))
    systolic_bp = get_numeric_input('Systolic Blood Pressure: ')
    diastolic_bp = get_numeric_input('Diastolic Blood Pressure: ')
    hip = get_numeric_input('Hip size in inches: ')
    waist_per_hip = waist / hip
    cholesterol_per_HDL = cholesterol / HDL

    # Forming a dataframe for the model
    data = {
        'Age_square': [age_sqr],
        'Square_Waist/Age': [sqr_waist_per_age],
        'GWC': [GWC],
        'Glucose': [glucose],
        'BMI': [BMI],
        'Chol/HDL ratio': [cholesterol_per_HDL],
        'Age': [age],
        'Cholesterol': [cholesterol],
        'HDL Chol': [HDL],
        'Gender': [gender],
        'Height': [height],
        'Weight': [weight],
        'Systolic BP': [systolic_bp],
        'Diastolic BP': [diastolic_bp],
        'waist': [waist],
        'hip': [hip],
        'Waist/hip ratio': [waist_per_hip]
    }

    df = pd.DataFrame(data)

    prediction = my_model.predict(df)

    if prediction == 1:
        return "This patient is likely to be diabetic."
    else:
        return "This patient is likely to be undiabetic."



result = predict_diabetes()
print(result)

Age of patient: 43
Waist size in Inches: 41
Weight of patient in lbs: 200
Fasting glucose level in mg/dL: 127
Cholesterol level in mg/dL: 293
HDL Cholesterol level in mg/dL: 60
Gender (Input '1' for male or '0' for female): 1
Height of patient in inches: 65
Systolic Blood Pressure: 141
Diastolic Blood Pressure: 97
Hip size in inches: 39
This patient is likely to be undiabetic.
In [ ]:
result = predict_diabetes()
print(result)
In [ ]: